Focus: AI and the accountant of the future

AI AND ACCOUNTING SOLUTIONS

The profession is experiencing its biggest skills shift in a generation. With AI now doing much of the routine work, accountants can no longer rely on process-heavy tasks to build competence. Instead, they need sharper judgement, stronger analytical thinking and the confidence to challenge machine outputs. As Stuart Pedley-Smith, Head of Learning at Synapse Learning Solutions argued, when information is cheap and automated tools are everywhere, the real differentiator becomes what people choose to do with that information. This means universities must rethink how they teach, what they assess, and how they prepare students for an environment where AI is embedded in everyday work from day one.

Pedley-Smith outlined the scale of the shift now underway, noting that large language models have made AI accessible to everyone, and that accessibility is transforming the nature of accounting work.

‘AI can now automate routine accounting tasks,’ he said. ‘That shifts the competencies the profession has traditionally relied on. Skills that once required years of training are now generated in seconds. Knowledge is free, it’s everywhere, and it’s updating faster than any syllabus.’

REWRITING THE COMPETENCY MAP

Pedley-Smith explained that as automation increases, the role of the accountant moves firmly toward judgement, interpretation, oversight and strategic thinking. The profession is shifting away from memorisation and towards the ability to ask the right questions of machines, evaluate responses, and understand when an AI-generated answer is wrong.

New learning opportunities are becoming possible, including adaptive learning pathways, AI tutors, personalised feedback and simulations that let students practise reasoning in real scenarios. ‘These tools expand human intelligence rather than replace it,’ he noted.

BUT ETHICAL AND PEDAGOGICAL RISKS CAN’T BE IGNORED

Pedley-Smith warned of ‘cognitive offload’ – the natural temptation for students to let AI think on their behalf. Other risks included unclear authorship, weaker critical thinking, misinformation, bias, and widening digital divides.

He stressed that tomorrow’s accountants will need to speak the language of AI, give models precise instructions and keep learning as the tools change. ‘Human-in-the-loop oversight is non-negotiable. Professional scepticism is not going anywhere.’

In the not-too-distant future, he said:

• AI marking with oversight becomes normal
• Adaptive assessment becomes mainstream
• Simulations and tutor avatars become standard
• Agentic AI will eventually handle multi-step accounting tasks
• Blended learning becomes the norm

AI FOR ACCOUNTING EDUCATION

If Pedley-Smith showed where the profession is
heading, the co-founders of Audit AI, Hackeem Dante Hafkey and Clint McLean, showed where students already are.

‘Students are using AI whether we approve of it or not,’ Hafkey said. Their research team at Audit AI monitors how students interact with different tools, and the findings were described as ‘uncomfortable but necessary’.

Students are turning to AI for:

• Explanations and concept breakdowns
• Summaries and research
• Walkthroughs of past papers
• Step-by-step study plans
• Practice questions

One student summed it up in their research notes: ‘AI doesn’t replace my studying. It just explains things the way I wish lecturers would.’

A WIDENING DIGITAL DIVIDE

McLean pointed to one of the most immediate equity risks created by AI. Not all students are using the same tools, and the gap is widening fast. ‘Wealthier students are paying for premium models that reason better and make fewer mistakes. Everyone else is stuck with the free versions.’

This affects the accuracy of explanations, the depth of reasoning, and the ability to solve multi-step accounting problems – the skills assessments are meant to measure.

‘If we do not address this divide, it will replicate and reinforce existing inequalities in the profession,’ McLean said. ‘AI access will soon be as essential as access to textbooks, devices or broadband, and universities need plans in place now.’

STUDENTS ARE EVOLVING INTO MULTI-TOOL USERS

Audit AI employs students as research assistants to stress-test different models. What surprised the team is how quickly students advanced. Within a few weeks, most moved from experimenting with a single AI tool to building layered workflows that mix several systems at once. They began combining language models, visual reasoning tools, step-by-step logic engines, music-based memory techniques and small applications generated through natural-language coding.

‘We didn’t teach them to do this,’ Hafkey said. ‘They just figured it out.’

Their outputs have become increasingly inventive. Some students used AI to generate music tracks that encode IFRS standards for easier recall. Others used ‘vibe coding’, where non-coders build functioning IFRS tools by describing what they want in plain language. A few created subject-specific tutor agents that outperform generalist AI models on technical tasks like tax calculations, audit assertions or disclosure analysis. It’s obvious that students have already embraced AI and that education now needs to catch up.

What this means for universities:

• Institutions need clear AI policies
• Students need guidance before bad habits form
• Lecturers require upskilling to understand how different models behave
• Access must be equitable
• Ethical AI behaviour needs to be taught early

WHY CONSISTENT AI GRADING IS HARD

Richard Mellon and Nic Pullen shifted the discussion to assessments, focusing on what AI-enabled grading can and can’t do. ‘AI isn’t a silver bullet,’ Mellon said. ‘It only creates new challenges.’

They explained that while AI can speed up marking and improve consistency, it fundamentally changes the work. Traditional marking is slow and fragmented, with multiple markers writing rubrics, rounds of moderation, endless sampling, late-night WhatsApp groups and the inevitable drift in judgement as people get tired.

AI flips that process. Rubrics are set once, applied instantly and adjusted through rapid test batches. Their workflow can be summarised as ‘click, grade, moderate, integrate’.

The result is faster marking and cleaner logic, but the presenters stressed that this does not remove the need for scepticism or human oversight. AI requires new skills: designing system prompts, managing inference, handling ambiguous answers, preventing hallucinations and knowing when human judgement must override the model.

They noted that the real value is not automation for its own sake, but freeing educators to focus on improving assessment quality, giving quicker feedback and spending more time on teaching rather than administration.

The result?

• 25× faster grading
• 70% cost savings
• More consistent logic
• Much faster feedback for students

Taryn Miller, Associate Professor at UT’s college of Accounting, said she was surprised by what she saw in the draft feedback from the pilot testing. ‘I realised the tool was actually marking better. It picked up logic that had sometimes been missed.’

Mellon and Pullen said AI grading requires new expertise:

• Designing strong system prompts
• Controlling inference
• Preventing hallucinations
• Handling edge cases
• Managing context windows
• Building quality assurance processes
• Choosing between reasoning models and standard LLMs

WHY GRADING MATTERS FOR THE FUTURE OF THE PROFESSION

They argued that speed is not the main benefit. The real value lies in what AI frees academics to do. With instant marking, lecturers can focus on −

• Giving more personalised and timely feedback
• Designing more authentic learning tasks spending more time on mentorship and developing professional judgement

Mellon and Pullen also noted that the same underlying technology can support interview simulations, candidate screening, mentor matching and workplace training, making it useful far beyond grading.

‘New world, new skills. The only way to learn them is by doing.’

Author
Monique Verduyn